Performance analysis of methods that overcome false sharing effects in software DSMs

Page-based software DSMs experience high degrees of false sharingespecially in irregular applications with fine grain sharinggranularity. The overheads due to false sharing is considered to be adominant factor limiting the performance of software DSMs. Severalapproaches have been proposed in the literature to reduce/eliminatefalse sharing. In this paper, we evaluate two of these approaches,viz., the Multiple Writer approach and the emulated fine grain sharing(EmFiGS) approach. Our evaluation strategy is two pronged. First, weuse an implementation-independent analysis that uses overhead counts tocompare the different approaches. Our analysis show that the benefitsgained by eliminating false sharing are far outweighed by theperformance penalty incurred due to the reduced exploitation of spatiallocality in the EmFiGS approach. As a consequence, any implementationof the EmFiGS approach is likely to perform significantly worse thanthe Multiple Writer approach. Second, we use experimental evaluation tovalidate and complement our analysis. The experimental results matchwell with our analysis. Also the execution times of the applicationfollow the same trend as in our analysis, reinforcing our conclusions.More specifically, the performance of the EmFiGS approach issignificantly worse, by a factor of 1.5 to as much as 90 times,compared to the Multiple Writer approach. In many cases, the EmFiGSapproach performs worse than even a single writer lazy release protocolwhich experiences very high overheads due to false sharing. The performance of the EmFiGS approach remains worse than the MultipleWriter approach even after incorporating Tapeworm-a record and replaytechnique that fetches pages ahead of demand in an aggregatedfashion-to alleviate the spatial locality effect. We next present theeffect of asynchronous message handling on the performance of differentmethods. Finally, we investigate the inter-play between spatiallocality exploitation and false sharing elimination with varyingsharing granularities in the EmFiGS approach and report the tradeoffs